GEOrca Index Report · Inaugural edition · July 2026

Who does AI actually cite?

Every day, the GEOrca Index asks six AI engines - Google AI Overview, Gemini, ChatGPT, Qwen and Kimi among them - the same 105 buyer questions across seven industries, in English and Chinese, and records every source their answers cite. This is what the first collection window (1,791 citations, snapshot 2026-07-03) shows.

Download the chart (PNG, free to republish with attribution) · Download the full dataset (CSV, CC BY 4.0)

1 · AI answers are built on UGC, not brand sites

YouTube or Reddit is the single most-cited source in every one of the seven categories we track - from GEO tools to freeze-dried food. Together they take 15.4% of all citations; add Quora, Zhihu, LinkedIn, Facebook and Medium and user-generated platforms carry 21% of everything AI cites. If your category's buyers ask AI for recommendations, the answer is being assembled from videos, threads and community posts - not from your homepage.

2 · The long tail is wide open

The top 10 domains take just 22.9% of citations - and 41.6% of all cited domains were cited exactly once. AI answers pull from a far wider pool than page-one Google, which means mid-size and niche sites get cited daily. The door is open; most brands simply haven't optimized for it.

3 · Wikipedia barely features

The most counterintuitive result: for commercial, buyer-intent questions, Wikipedia earned under 1% of citations (16 of 1,791). AI engines reach for comparisons, reviews, community threads and vendor content - not encyclopedias - when someone asks "which tool should I use?".

4 · Every engine has its own taste

Google AI Overview cites densely (it grounds on live search results). Qwen (通义) leans on Chinese platforms like Zhihu. Chat-API models without web grounding (ChatGPT's API tier, Kimi) rarely cite full URLs at all - they mention brands instead. Practical consequence: AI visibility is a per-engine game; being cited by one engine says nothing about the others.

The top 20 most-cited domains

What the research says

Our findings line up with the published research on AI search behavior:

"Google users who encounter an AI summary are less likely to click on links to other websites than users who do not see one" - clicking a traditional result happened in just 8% of visits with an AI summary vs 15% without, and links inside the summaries were clicked in only 1% of visits.

Peer-reviewed research on Generative Engine Optimization found that GEO methods can improve a source's visibility in generative-engine responses by "up to 40%".

Pew also found Wikipedia, YouTube and Reddit to be the most-cited sources across all Google AI summaries. Our buyer-intent question set shows the same UGC dominance for YouTube and Reddit - but Wikipedia all but disappears when the question is commercial ("which tool should I buy"), which is exactly the slice that matters for brands.

Methodology

• • 105 buyer-intent questions across 7 industries (GEO/AI-search tools, e-commerce, finance, B2B trade, cashback, affiliate marketing, freeze-dried foods), in English and Chinese.
• • Asked daily to Google AI Overview, Gemini, ChatGPT (API), Qwen and Kimi; every cited source recorded.
• • Snapshot: collection window ending 2026-07-03; 1,791 citations, 688 domains. GEOrca's own brand queries are excluded.
• • The live, daily-updated leaderboard is public: georca.com/index.

← GEOrca Index · About the Index · 中文